A technology that can identify individuals from the GPU in use has appeared, and even GPUs of the same model number can be accurately distinguished

Advertisements on the Internet track user behavior and select appropriate targets in order to maximize their effectiveness. This user tracking is done using cookies etc., but a new technology that can identify users using GPU was announced.

DRAWNAPART: A Device Identification Technique based on Remote GPU Fingerprinting
https://arxiv.org/abs/2201.09956

Researchers use GPU fingerprinting to track users online
https://www.bleepingcomputer.com/news/security/researchers-use-gpu-fingerprinting-to-track-users-online/

In targeting advertising , which analyzes user behavior and distributes advertisements, it was common to use third-party cookies to identify individuals, but in recent years, multiple browsers have begun to block third-party cookies. Therefore, the advertising industry needs to establish a new personal identification method. Among the new personal identification methods, a method to identify an individual by creating a 'fingerprint' based on 'OS', 'browser', 'time zone', 'hardware configuration', 'network setting', etc. It is attracting attention because of its ability to accurately identify individuals. You can understand more about fingerprints by reading the following article.

What is a 'fingerprint' that can identify users without cookies? – GIGAZINE.BIZ

As mentioned above, the user's hardware configuration is used to create the fingerprint, and this hardware configuration includes information such as 'screen resolution', 'free storage space', 'number of CPU cores', and 'number of cameras and microphones'. is used. The newly announced personal identification method using GPU is named ' DRAWNAPART ', and it is possible to further improve accuracy by using it together with the personal identification method using the above information.

The processing required for DRAWNAPART is to let the user's GPU execute arbitrary calculations using ' WebGL ', a technology that allows GPU processing to be performed on the browser, and to measure the time required for the calculation. Of course, the time required for calculation processing varies depending on the type of GPU, but due to the manufacturing process of GPUs, there are subtle differences in performance even between GPUs of the same type. It is possible to identify an individual.

According to the research team that published DRAWNAPART, individual identification using conventional fingerprints alone could only track individuals for an average of 17.5 days, compared to an average of 28 days using conventional fingerprints and DRAWNAPART. I was able to track it for days. However, since the above tracking days measurement is performed under the conditions of ``keeping the temperature of the GPU between 26.4 degrees and 37 degrees'' and ``no voltage fluctuation'', tracking may be difficult if the temperature or voltage changes. there is.

In the research report, the research team cites that `` WebGPU '' is under development as a successor technology to WebGL, and claims that ``If WebGPU becomes common, it is expected that a personal identification method similar to DRAWNAPART will appear.'' doing.